OptiFlow — AI-Enhanced Retail Inventory Monitoring
Built an AI-enhanced system combining wearable RFID scanning with UWB indoor positioning for Decathlon, enabling real-time shelf-level inventory monitoring and store flow optimization.
- ▸Custom wearable hardware: ESP32-S3 microcontroller + UHF RFID reader + UWB positioning module
- ▸Real-time web dashboard with store maps, employee trajectories, and stock heatmaps
- ▸ML pipeline for product clustering, demand forecasting, anomaly detection, and ABC classification
- ▸Best in class — top marks, client was extremely happy with the solution
Overview
From small retailers to companies like Decathlon, everyone deals with the same problem: what's supposed to be on the shelf often isn't, and figuring out where your inventory actually is in real-time is harder than it sounds. Existing solutions either don't scale, aren't real-time, or raise privacy concerns.
OptiFlow tackles this with a system that turns routine employee movement into a continuous sensing layer — no cameras, no manual scanning.
How It Works
The core idea: sales associates wear a small custom device while doing their normal tasks. As they walk around the store, the device passively scans RFID-tagged products and tracks its own position using UWB, building a real-time picture of what's on which shelf.
Hardware
We built a custom wearable platform combining:
- ESP32-S3 microcontroller running multi-tasking firmware
- JRD-100 UHF RFID reader for scanning product tags
- DWM3001CDK UWB module for centimeter-level indoor positioning
- A network of stationary UWB anchors deployed throughout the store
The firmware handles concurrent sensor polling, data collection, and wireless communication via MQTT — all running in real-time on a resource-constrained device.
Software
A containerized microservice architecture processes the incoming telemetry:
- Real-time trilateration from UWB distance measurements
- Dual-mode missing item detection — comparing expected vs. observed inventory
- ML analytics pipeline — product clustering, demand forecasting, anomaly detection, and ABC classification
- Next.js dashboard showing live store maps, employee trajectories, stock heatmaps, and admin controls
Business Viability
We also did financial modeling across three deployment scenarios — pilot store, full store, and scalable UL-TDOA architecture — showing that the system is commercially viable through a hybrid hardware + SaaS revenue model.
Results
Decathlon was extremely happy with what we delivered. The project was rated best in class and received top marks. We demonstrated that continuous, real-time shelf monitoring is both technically feasible and financially sustainable — without compromising privacy or requiring heavy infrastructure.